consul-resolver

A simple consul library for load balancing and metrics tracking with redis. This package provides multiple load balancing algorithms, such as Round Robin, Least Connection, and Weighted Round Robin.

Features

• Multiple load balancing algorithms:
  • Round Robin
  • Least Connection
  • Weighted Round Robin
• Consul Integration

Installation

yarn add consul-resolver

npm install consul-resolver

Prerequisites

• Node.js >= 14
• Redis server
• Consul server

Quick Start

import { ConsulResolver, SelectionAlgorithm } from "consul-resolver";
import Redis from "ioredis";
import https from "https";

const redis = new Redis({
  host: "localhost",
  port: 6379,
});

const resolver = new ConsulResolver({
  redis,
  host: "127.0.0.1",
  port: 8500,
  secure: false,
  cachePrefix: "mydb",
  token: process.env.CONSUL_TOKEN,
  agent: new https.Agent({
    rejectUnauthorized: false,
    minVersion: "TLSv1.2",
    maxVersion: "TLSv1.3",
  }),
});

const service = await resolver.selectOptimalService(
  "my-service",
  SelectionAlgorithm.LeastConnection,
);

await resolver.incrementConnections(service.selected.id);

await resolver.decrementConnections(service.selected.id);

Configuration

The ConsulResolver constructor accepts the following configuration options:

interface ConsulResolverConfig {
  cachePrefix: string; // Prefix for Redis cache keys
  redis: Redis; // Redis instance
  host: string; // Consul host
  port: number; // Consul port
  secure: boolean; // Use HTTPS
  token?: string; // Consul ACL token
  agent?: Agent; // Optional HTTPS agent configuration
  weights?: {
    health: number;
    responseTime: number;
    errorRate: number;
    resources: number;
    connections: number;
    distribution: number;
  }; // Custom weights for the weighted round robin algorithm
  metrics?: {
    responseTime: number;
    errorRate: number;
    cpuUsage: number;
    memoryUsage: number;
    activeConnections: number;
  }; // Custom metrics for the weighted round robin algorithm
}

API Reference

selectOptimalService(service: string, algorithm?: SelectionAlgorithm): Promise<OptimalServiceResult | null>

Selects the optimal service instance based on the specified algorithm.

getSelectionMetrics(serviceId: string): Promise<ServiceMetrics | null>

Retrieves current metrics for a specific service.

incrementConnections(serviceId: string): Promise<void>

Increments the active connection count for a service.

decrementConnections(serviceId: string): Promise<void>

Decrements the active connection count for a service.

refresh(): Promise<void>

Clears all stored metrics from Redis.

Types

OptimalServiceResult

interface OptimalServiceResult {
  selected: {
    ip: string;
    port: number;
  } | null;
  services: Array<{
    ip: string;
    port: number;
  }>;
}

ServiceMetrics

interface ServiceMetrics {
  responseTime: number;
  errorRate: number;
  cpuUsage: number;
  memoryUsage: number;
  activeConnections: number;
  lastSelectedTime?: number;
}

Service Selection Weights and Metrics

Selection Weights

The Weighted Round Robin algorithm uses the following weight distribution to calculate service scores:

const DEFAULT_WEIGHTS = {
  health: 0.25, // Service health status (25%)
  responseTime: 0.2, // Response time performance (20%)
  errorRate: 0.2, // Error rate of the service (20%)
  resources: 0.15, // CPU and memory usage (15%)
  connections: 0.1, // Active connection count (10%)
  distribution: 0.1, // Time since last selection (10%)
};

Weight Explanations

• health (25%): Prioritizes services with passing health checks

  • Calculated as ratio of passing checks to total checks
  • Most heavily weighted as service health is critical

• responseTime (20%): Favors services with lower response times

  • Normalized against a 500ms baseline
  • Higher weight indicates better performance

• errorRate (20%): Considers service reliability

  • Normalized against a 100% scale
  • Lower error rates result in higher scores

• resources (15%): Accounts for service load

  • Combines CPU and memory utilization
  • Prevents overloading of busy instances

• connections (10%): Active connection count

  • Helps distribute load across instances
  • Prevents any single instance from being overwhelmed

• distribution (10%): Time since last selection

  • Ensures fair rotation among services
  • Prevents "hot spot" instances

Default Metrics

Each service starts with these default metrics if no historical data is available:

const DEFAULT_METRICS = {
  responseTime: 100, // Default 100ms response time
  errorRate: 0, // Start with 0% error rate
  cpuUsage: 50, // Assume 50% CPU usage
  memoryUsage: 50, // Assume 50% memory usage
  activeConnections: 0, // Start with no active connections
};

Metrics Explanation

• responseTime: Initial 100ms baseline

  • Conservative default for new services
  • Updated based on actual performance

• errorRate: Starts at 0%

  • Optimistic initial error rate
  • Adjusted based on actual failures

• cpuUsage: Default 50%

  • Moderate initial CPU load assumption
  • Updated with actual metrics when available

• memoryUsage: Default 50%

  • Moderate initial memory usage assumption
  • Updated with actual metrics when available

• activeConnections: Starts at 0

  • Fresh services begin with no connections
  • Incremented/decremented as connections are established/closed

Usage with Express (Best used as a middleware)

import express from "express";
import { ConsulResolver, SelectionAlgorithm } from "consul-resolver";

const app = express();
const resolver = new ConsulResolver(config);

app.use(async (req, res, next) => {
  const service = await resolver.selectOptimalService("api-service");
  if (!service) {
    return res.status(503).json({ error: "No service available" });
  }

  await resolver.incrementConnections(service.selected.id);

  req.serviceInfo = service.selected;

  res.on("close", async () => {
    await resolver.decrementConnections(service.selected.id);
  });

  next();
});

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT

Author

Muritala David Ilerioluwa